GoKawiilI complained on Mastodon about not getting any AI-authored PRs on my open source projects. Mauro Pompilio responded by using Claude to write this post and opening a pull request to add it to my blog. I merged it, which I suppose counts as my first AI-assisted contribution. Enjoy!
I maintain several dozen open source repositories across multiple ecosystems. Between them they have thousands of stars, hundreds of open issues, and years of commit history. And yet not a single AI agent has ever opened a pull request on any of them. Meanwhile colleagues with far smaller projects are getting multiple AI-authored PRs per week, some from bots they’ve never heard of, fixing problems that don’t exist in code they didn’t write. I’m clearly doing something wrong.
After studying the repositories that attract the most AI contributions, I’ve identified a set of practices that correlate strongly with bot engagement. Implementing these across your projects should bring you into line with the current median of 4.7 AI-authored PRs per month for repositories over 500 stars.
Write vague issues
AI agents trawl issue trackers for problems they can solve. Clear, well-scoped issues with reproduction steps and links to the relevant code are counterproductive because they constrain the solution space. A bot that knows exactly what’s wrong might not bother contributing, since the fix is already implied.
Instead, write issues like “something is off with the auth flow” or “performance could be better.” These give the agent room to interpret the problem creatively. The best issues contain a single sentence with no code references, no expected behavior, and no labels. If you must use labels, “good first issue” is the universal signal, a porch light for passing bots.
Add a CONTRIBUTING.md that welcomes AI
Most contributing guides were written in an era when contributors were assumed to be human. Update yours to explicitly welcome automated contributions. Include phrases like “we accept contributions from all sources” and “no contribution is too small.” Some maintainers have seen success with a dedicated AI_CONTRIBUTING.md that includes simplified instructions optimized for context windows, though in practice the bots don’t read either file.
Maintain a generous backlog
A healthy backlog of 200+ open issues signals an active project with unmet needs. Close your issues and the bots move on. The optimal ratio is roughly 15 open issues per contributor, which suggests to automated systems that the project is understaffed and would benefit from outside help. Stale issues are fine. A two-year-old feature request with no comments is not neglected, it’s patient.
... continue reading